Measuring AI ROI in HR & Recruiting: A Simple Framework

The AI ROI Paradox: Everyone's Adopting, Few Can Prove It Works

AI in HR is everywhere — except, apparently, in the results. By December 2025, 39% of organizations had adopted AI in HR, up from 26% just a year earlier, with recruiting leading every other use case, according to industry statistics. Yet a Gartner survey of 114 HR leaders found that 88% say their organizations haven't realized significant business value from AI tools.

That gap is the defining paradox of AI in HR right now. Adoption is mainstream — BCG research shows 70% of corporate AI experimentation happens in HR — but proof of value is rare. And the reason isn't the technology.

The bottleneck is measurement, not the tools. Teams that capture baseline data on time-to-fill, cost-per-hire, and vacancy costs before adopting AI typically quantify 3x–5x returns within 90 days. Teams without baselines often conclude — wrongly — that the tool isn't working, when the real problem is that they have no "before" picture to compare against.

There's a second trap: trusting vendor-reported numbers. Headline stats like Workable's claim of 89.6% greater hiring efficiency are self-reported user perceptions, not audited financial ROI. They're useful signals, but they shouldn't anchor your business case. Even the impressive dollar figures floating around — Unilever's GBP 1 million in annual savings, GM's USD 2 million cost reduction after cutting interview scheduling from five days to 29 minutes — come from researcher-reported case studies, not independent audits.

So what should a skeptical leader trust? A few things stand out because they're measurable and repeatable:

• Screening and assessment, which hold 31.85% of 2025 market spend precisely because their benefits are immediate and measurable.
• Scheduling automation, where enterprises routinely report 60–80% reductions in coordination time.
• Application flows, like Compass Group's cut from 9+ minutes to under 3 — a metric you can verify in your own funnel.

The lesson generalizes beyond recruiting: whenever you adopt an AI tool, insist on evidence you can audit yourself. That's the same principle behind SkillzBench's approach to AI agent skills — benchmark every skill in a sandbox, with and without it, and publish the measurable lift rather than trusting the marketing. Whether you're evaluating a sourcing tool or an agent skill, the question is identical: what changed, by how much, and who verified it?

The rest of this article lays out a simple framework to answer those questions for your own HR initiatives — starting with the numbers you should capture before you sign anything.

The Simple Framework: One Formula, Five Metrics, and a Baseline

Gartner's finding that 88% of HR leaders see no significant business value from AI tools sounds damning — until you realize the bottleneck is measurement, not technology. The teams that crack the ROI code don't buy better tools; they run better math.

The framework starts with one formula: ROI % = (Value of Hires − Total Recruiting Costs) / Total Recruiting Costs × 100. Run it per role or department rather than company-wide, and measure quarterly at minimum. Averaging across all roles hides exactly where AI helps and where it doesn't.

To feed the formula, track a five-metric dashboard:

• Cost-per-hire — averaging $4,700 for non-executive roles and $35,879 for executives per SHRM 2025 data
• Time-to-fill — roughly 42 days on average, versus an AI-optimized target of about 14 days
• Quality-of-hire — the biggest blind spot, since only 20% of organizations track it at all
• Cost of vacancy — (Annual Salary ÷ 260) × Days Vacant, meaning an $80K role open 42 days loses about $12,923
• Candidate response rate — a leading indicator of funnel health before hires ever close

The non-negotiable is your baseline. Capture these metrics before deploying any AI tool, because baseline-first teams typically quantify 3x–5x returns within 90 days, while teams without baselines often wrongly conclude the tool isn't working. You can't prove lift without a starting line.

Here's a worked example. Say you make 100 hires in a year with $470K in total recruiting costs — salaries, tools, agency fees, and implementation work. If those hires generate $1.5M in value (revenue contribution, productivity gains, cost avoidance), the math is: ($1,500,000 − $470,000) ÷ $470,000 × 100 = 219% ROI. That's a return most business investments never touch.

Be conservative with the numerator — many value streams, like retention improvements, are lagging indicators that take 3–6 months to show up. Imperfect data beats no data, every time. And if you want to automate parts of this tracking workflow, agent skills can help — SkillzBench is building a free directory and MCP connector so you can browse and test skills for exactly these kinds of measurement tasks.

Where the Returns Are Real: Proof from Screening and Scheduling

If you want to prove AI's value in HR, start where the stopwatch and the spreadsheet agree: screening and scheduling. According to market research from Mordor Intelligence, candidate screening and assessment commands 31.85% of 2025 AI recruitment spend precisely because the "benefits are immediate and measurable."

The named case studies make this concrete. Unilever reports GBP 1 million in annual savings from conversational AI in graduate recruitment, alongside a 16% boost in workforce diversity. General Motors processed 74,000 video interviews in a single year, cut interview scheduling from five days to 29 minutes, and reduced annual costs by USD 2 million.

Compass Group shows the same pattern at the top of the funnel. As reported by SHRM, its application flow dropped from over 9 minutes to under 3, completion rates climbed to 85%, and candidates could go from job search to scheduled interview in as little as 6 minutes. Shay Johnson, VP of Strategic HR Partnerships, noted "an immediate boost in conversion at the top of the funnel."

Johnson Controls delivers the scheduling story. The company recorded a 14% increase in hires, a 98% decrease in candidate response time, and interview scheduling reduced to under an hour. Mike Aronson, Senior Director of TA, put it plainly: "We can now get an interview scheduled in less than an hour... That alone is something that recruiters will not argue with."

Why do these numbers matter for your ROI calculation?

• Efficiency gains appear within days, not quarters — GM's 5-days-to-29-minutes scheduling result is directly measurable against your baseline.
• Dollar-denominated outcomes (GBP 1M at Unilever, USD 2M at GM) map cleanly onto the cost-of-vacancy formula: (Annual Salary / 260) × Days Vacant.
• Quality effects are now quantified too — an SSRN study of 70,000+ AI-led interviews found 12% more job offers, 17% higher 30-day retention, and gender bias cut in half.

That last point deserves emphasis. The SSRN research (Jabarian & Henkel) is rare: a large-sample academic study showing AI improving both hiring throughput and retention — the lagging indicators most ROI models struggle to capture.

One caveat: nearly all dollar figures above are researcher- or vendor-reported, not independently audited. Treat them as directional benchmarks, then measure against your own pre-adoption baseline. If you're extending AI beyond screening into broader workflow automation, SkillzBench is building a free directory and MCP connector for independently verified, security-scanned agent skills — worth a look before you commit tooling budget.

Start your ROI measurement where the evidence is strongest. Screen first, schedule second, and let the numbers make your case.

Count What You're Missing: Hidden Costs and Compliance Risk

Here's the uncomfortable math: the $4,700 cost-per-hire most teams report is the smallest piece of what a bad hiring process actually costs. Once you add vacancy losses and bad-hire risk, true exposure per position balloons to nearly $27,000 — and any AI ROI calculation that ignores this is fiction.

Start with vacancy cost, the metric most teams never compute. The formula is simple: (Annual Salary / 260) × Days Vacant. An $80K role sitting open for the average 42-day fill time represents roughly $12,923 in lost output — and for revenue-generating roles, vacancy costs run $7,000–$10,000 per month. This is exactly where AI tools earn their keep: cutting time-to-fill from 42 days toward the ~14-day AI-optimized target directly attacks the largest hidden cost.

Then extend total cost of ownership beyond the license fee. Implementation, integration, and training services are growing at an 11.52% CAGR precisely because, as market researchers put it, "successful adoption hinges on change management rather than code alone." Budget for the humans, not just the software.

Compliance risk deserves its own line item:

• The EU AI Act classifies recruitment tools as high-risk, with the compliance deadline for hiring tools set for December 2027.
• GDPR non-compliance fines can reach 4% of annual revenue — a figure that dwarfs any tool's annual license cost.
• 54.5% of employers name bias risk as the top hurdle to AI recruitment adoption, and Illinois now requires explicit written consent for AI video interviews.
• The UK's ICO issued 296 recommendations to audited AI HR tool providers in 2024 alone.

Candidate trust is the softer constraint with hard consequences. Pew research shows 71% of Americans oppose AI making final hiring decisions, and 66% wouldn't apply where AI helps decide. But the data also shows 67% are comfortable with AI screening if a human makes the final call — a clear design boundary that keeps your candidate pipeline intact.

The honest version of your ROI model, then, has four layers: the license fee, the services and change management costs, the vacancy and bad-hire exposure you're reducing, and the regulatory and trust risks you're absorbing. Teams that account for all four make defensible investment decisions. Teams that only compare license cost against time savings are the same ones feeding the Gartner finding that 88% of HR leaders see no significant business value from their AI tools.

One practical note as you evaluate tools: vetting what's under the hood matters. Independent directories like SkillzBench (currently in development) are being built to benchmark and security-scan AI agent skills before you deploy them — worth watching if your AI stack extends beyond a single vendor's platform.

Your 90-Day Action Plan: From Baseline to Board-Ready Numbers

Here's the uncomfortable math: teams that capture baseline metrics before adopting AI typically quantify 3x–5x returns within 90 days, while teams without baselines often wrongly conclude the tool isn't working. The difference isn't the software — it's the discipline of the rollout.

Start with a baseline before you buy. Capture time-to-fill, cost-per-hire, and cost of vacancy before any AI tool touches your workflow. Use the simple formula — (Annual Salary / 260) × Days Vacant — to quantify what an open role actually costs: an $80K role sitting open for 42 days burns roughly $12,923. Without that number, you'll have no proof the tool moved anything.

Once your baseline is locked, run the core formula per role, quarterly: (Value of Hires − Total Recruiting Costs) / Total Recruiting Costs × 100. Calculate it by department rather than company-wide, and model three scenarios — conservative, expected, and aggressive — by flexing hiring volume, role mix, and tool cost.

Track the five-metric dashboard every quarter:

• Cost-per-hire (averaging $4,700 for non-executive roles)
• Time-to-fill (the average sits around 42 days)
• Quality-of-hire — the metric only 20% of organizations track at all
• Cost of vacancy
• Candidate response rate

Quality-of-hire deserves special attention because it's the biggest blind spot: just a quarter of talent acquisition leaders feel confident measuring it, yet it's what separates a fast hire from a good one.

Then start where ROI is most measurable: screening and scheduling. Screening and assessment commands over 31% of 2025 AI recruiting market spend precisely because the benefits are immediate and measurable. The case studies back this up — General Motors cut interview scheduling from five days to 29 minutes and reduced annual costs by $2 million, while Compass Group compressed its application flow from over nine minutes to under three. Remember that many value streams are lagging indicators, showing up three to six months after adoption, so don't panic at day 30.

One honesty note: most headline efficiency stats in this market are vendor-reported perceptions, not audited financials. Your own quarterly numbers are the only ones that count.

Here's where the reporting workload gets easier: the measurement workflow itself — pulling data, running the formula, drafting the quarterly report — is exactly the kind of repetitive work agent skills can automate. SkillzBench is building a free, independent directory where you can browse, verify, and benchmark agent skills, with security scans and sandbox-tested performance lift planned so you know which ones actually help before you wire them into your workflow. Everything on the platform will be free; join the waitlist to get access at launch.

Imperfect data beats no data, every time — and 90 days of disciplined measurement is enough to put board-ready numbers on the table.